Exploring Object Stores for High-Energy Physics Data Storage
چکیده
Over the last two decades, ROOT TTree has been used for storing over one exabyte of High-Energy Physics (HEP) events. The columnar on-disk layout proved to be ideal analyses HEP data that typically require access many events, but only a subset information stored each them. Future colliders, and particularly HL-LHC, will bring an increase at least order magnitude in volume generated data. Therefore, use modern storage hardware, such as low-latency high-bandwidth NVMe devices distributed object stores, becomes more important. However, was not designed optimally exploit hardware may become bottleneck retrieval. RNTuple I/O system aims overcoming TTree’s limitations providing improved effciency systems. In this paper, we extend with backend uses Intel DAOS underlying storage, demonstrating architecture can accommodate high-performance stores. From user perspective, accessed minimal changes code, is by replacing filesystem path URI. Our performance evaluation shows new realistic analyses, while outperforming compatibility solution provided project.